The pharmaceutical industry is currently trapped in a mathematical decay known as Eroom’s Law. While Moore’s Law tracks the exponential rise of computing power, Eroom’s Law—its namesake spelled backward—describes the inverse reality of drug development: the cost of bringing a new drug to market has roughly doubled every nine years since the 1950s. Today, as Paul Belcher, director of protein research strategy at Cytiva, points out, the average timeline spans 10 to 15 years with a price tag between $1 billion and $2.5 billion. With failure rates upward of 90%, the industry is betting on AI to reverse this trend, yet the primary obstacle is no longer just digital—it is stubbornly physical.
The Predictive Shift and the Testing Bottleneck
AI is fundamentally changing the starting point of R&D by moving from empirical screening to predictive design. In the past, researchers physically screened massive libraries of molecules to find a 'hit'—a molecule that binds to a disease target. According to Belcher, AI now allows companies to design candidates from scratch and predict interactions before committing any physical resources. This removes the traditional limit on how many compounds a company can screen, but this efficiency creates a new friction point. We are seeing a massive surge of AI-generated candidates that all require laboratory validation, creating a high-speed collision between digital hypothesis and physical reality.
As Belcher explains, AI cannot yet reliably predict the kinetics or developability of new compounds. The result is a 'physical bottleneck' where lab teams, traditionally equipped to handle binary 'yes-or-no' data, are now overwhelmed by the need to characterize and purify a growing volume of diverse, complex candidates. The speed of digital generation has officially outpaced the throughput of the physical lab, turning the laboratory into the most expensive waiting room in the world.
Breaking the Data Wall
To survive, the industry must transition to a 'closed data loop' where physical results feed back into models in real-time. This high-fidelity, structured data is the only way to reduce risk in the clinical phase—the stage Belcher identifies as the primary cost driver in drug discovery. By identifying low-quality candidates earlier through predictive analytics rather than simple RAG-based systems, firms can finally compress timelines. The goal is to shift the structure of costs: spending more on high-quality data early on to avoid the catastrophic $100 million failures during late-stage clinical trials.
Investors and R&D leaders are often sold a vision where AI replaces the laboratory, but the reality is that AI has only made the laboratory more critical. The industry promised that algorithms would slash the entry price for new therapeutics. Instead, the immediate result is a desperate need for sophisticated, high-throughput physical infrastructure to handle the digital overflow. They promised a virtual shortcut; instead, they delivered a reality check: you cannot automate biology without mastering the hardware that measures it.